New generative AI technology never seems to emerge slowly onto the scene—it arrives with a bang.
OpenAI’s ChatGPT sent ripples around the world as people started to become aware of this new tech. Google’s Gemini AI had a difficult birth with some high-profile blunders. But none have been quite as seismic as the recent launch of DeepSeek.
DeepSeek’s arrival is important for several reasons. First, it costs around a tenth of the price of the other models to use, including Anthropic’s Claude, Mistral’s LLaMA and Meta AI along with the aforementioned models but it is about twice as fast.
Second, it doesn’t use (or shouldn’t at least, for reasons we’ll explain shortly) Nvidia’s microchips to power its operations. After DeepSeek emerged, the company lost nearly AU$1 trillion in market capitalisation, though not exclusively as a result of DeepSeek.
Third, and perhaps most importantly, DeepSeek is Chinese. Until now, the US has largely had a monopoly over AI innovators. DeepSeek, meanwhile, with its cut-price but high-speed operation has turned the game on its head. Anyone or any business serious about using generative AI should be using DeepSeek, right?
Proceed with Caution…
“I heard when they released the original model late last year but I didn’t pay too much attention to it,” Tim O’Neill, co-founder of generative AI agency Time Under Tension told B&T.
“But it’s really in the last week with all the noise around the reasoning model and also that they published the research to show how they’ve done it. The thing that’s really blown everyone away is how they’ve managed to build a really high quality model at a fraction of the cost that’s also twice the speed.”
O’Neill wasn’t alone in noticing the new model’s potential. Its iOS app rocketed to the top of the downloads chart in 51 countries, including the US and displacing ChatGPT in the process. On Friday, DeepSeek’s mobile app had 0nly a million downloads across both the App Store and Google Play. That number has since doubled as of Monday morning, to 2.6 million downloads of DeepSeek’s mobile app across both platforms.
In addition, more than 80 per cent of DeepSeek’s total mobile app downloads have come in the past seven days, according to analytics firm Sensor Tower. In that time frame, DeepSeek saw nearly 300 per cent more app downloads than Perplexity, another leading consumer AI app.
But while the public has been clamouring to get its hands on tech, pollies and experts from around the world have exercised caution.
Federal Science Minister Ed Husic said he was counselling national security agencies about DeepSeek’s potential threats. However, he cautioned that it was too early to ban the AI model like the previous coalition government banned Chinese telco Huawei from Australia’s 5G rollout.
“I think there’ll be parallels to what you’ve seen in the discussion around TikTok that emerge around DeepSeek as well,” he said.
“We’ll clearly be informed by the advice of the national intelligence community in relation to threats as they might present at different points in time. How do we respond? How do we shape up that response? We’ll take that on board and deal with it then.”
Tech Council of Australia (TCA) chief executive Damian Kassabgi issued more grave advice.
“Act now or risk Australia falling behind in AI development and adoption,” he said.
“DeepSeek’s reported breakthrough shows that the AI landscape is highly competitive and rapidly evolving. To ensure Australia can keep up, we need co-ordinated action supported by a national plan,”.
“We support the Government’s National AI Capability Plan, which we began calling for in August 2023, but this work needs to be brought forward. Australia cannot wait for the plan to be finalised by the end of 2025,” Kassabgi added.
“Development and adoption of AI in Australia could create up to $115 billion in economic value annually and 200,000 jobs by 2030, according to Tech Council research. But realising these benefits will require the right policy settings and co-ordination with industry to ensure Australia is a competitive place to make and deliver technology products.”
… Especially As A Business
“It’s amazing, but don’t use it,” added O’Neill.
“That’s my recommendation. It’s important and amazing but don’t use it.
“The first thing we tell our clients when they’re using ChatGPT or Gemini that you go into the settings and turn off ‘Train on my data’ because, by default, if you upload your brief for your agency or from your client, that will go into the training data and become basically publicly available. It’s no longer secret. So that’s bad.
“DeepSeek doesn’t have any of those settings, so anything you do will be used to train the model. It’s also going to China and they are reasonably clear about if they need to hand over any information to the Chinese Communist Party, then they’ll do that. Don’t put anything confidential into DeepSeek, is the bottom line.”
That your data, or worse, your client’s, could end up in the hands of the Chinese Communist Party should be enough to put the frighteners on any agency or business. Not that many are taking advantage of AI at all, however.
“Number one, don’t go and use DeepSeek for [just] anything. Go muck around with it for personal use, sure, but don’t use it for business. To me, though, maybe I’m biased, but there are still so many clients that can’t get their head around this stuff and not getting started. The boat is leaving the port. If companies and agencies aren’t already on it, someone else will be,” added O’Neill.
So tread carefully, but not too slowly, it seems. Another point worth mentioning, DeepSeek is not a multimodal AI tool—yet. As it stands, the model can only do text whereas others, including ChatGPT, can make you images, videos and more.
Winner Takes All
The other side to the DeepSeek conundrum, however, is how its entrance to the market and explosion in popularity affects usage of other AI technologies.
The world of search has, infamously, become an effective monopoly with Google controlling most of the market in most of the world.
Will the same happen with generative AI-powered searches? It isn’t out of the realms of possibility.
“Four weeks ago, the answer to that was there would have been one or two big winners… most likely OpenAI,” said O’Neill.
“They had a moat and that moat was technical plus data plus servers plus infrastructure plus money. The whole US$500 billion investment was their moat. There was no one else in the world that had the money to build the infrastructure needed to train these models, and the best model will win.
“That has been shattered. Now way more companies can compete on the technical side. They don’t need US$500 billion.”
Google and Meta, O’Neill said, are still very much in the race and their established brand presences and user bases will give them a significant advantage. But the battle for AI dominance has been blown wide open.